Research On Key Techniques Of Short-Term Traffic Forecasting Toward Vehicle Navigation System Application

In recent years, following the unceasing enhancement of city motorization level as well as the gross quantity of personal Car climbs rapidly, the Beijing Urban traffic demand has been growing continually, but the land resource, the traffic resources supplies in the urban are limited; therefore, the conflict between road supply and demand become more and more prominent. In view of the present situation and the existing problems of Beijing road traffic, in particular for the demand of urban traffic control during the period of the Beijing Olympic Games 2008, the comprehensive promotion and widespread application of the Intelligent Transportation System (ITS) in urban transportation management have the prominent significance.The Dynamic Vehicle Navigation System (DVNS) is one of the most important application subsystems which service for the travelers under the framework of Intelligence Transportation System (ITS). It has the advantages on balancing the traffic flow in the road network, alleviating the urban traffic jam. But the data collection, processing as well as the distribution of real-time urban traffic information are the foundation to show the function of DVNS; at the same time, it's also the only way to break the develop bottleneck of Static Vehicle Navigation System (SVNS) in China.The research of this Dissertation depends on the project of the transportation guidance (vehicles navigation) system technology and the equipments research under the Beijing Intelligence Transportation System (ITS) demonstration project. The research takes the accurate forecast and distribution of traffic information as a goal, and focuses on the data processing technologies regarding the Dynamic Vehicle Navigation System (DVNS), including the research on the Floating Car (FC) based dynamic traffic information processing algorithm, multi-source historical traffic condition data processing technologies as well as the study on the historical database based short-term traffic forecasting model.Dynamic traffic information collection is the key to realize the traffic condition short-term forecasting. The paper studied and developed GPS based Floating Vehicle Data Collection System (FVDCS). Then, the paper studied the processing techniques of collected Floating Car Data (FCD) with emphasis. Through the comprehensive analysis of the original FCD, base on the traffic engineering theory, the paper proposed stay time estimating method based FCD process algorithm. It can accurately obtain the latest travel speed information of the city road network. The field experiment results showed that, the average accuracy of algorithm is over 90 percents, so it indicates that the algorithm is capable of processing the FCD into travel speed simply and accurately. Through the comparison with other algorithms from domestic researches, it also indicated that the algorithm established in this article has the high accuracy and the good reliability.Historical traffic data processing and the preparation is the important foundation to realize the traffic condition short-term forecast. Paper classified the historical transportation condition data which collected via the FVDCS, and proposed the three-stage applications of FCD. The paper proposed the pretreatment technology on the aspects of data filteringand the data remedying, then, the heuristic soft threshold selection method based wavelet de-noising technology was developed to process the historical data series; at last, the system cluster and the K-mean cluster based dual cluster methods was used to condense data series. Thus, the research established the integrated historical databases which contained kinds of typical traffic evolution tendency of the various class roads.On the basic of the historical tendency database and real-time traffic information, this research aimed at the short-term traffic forecast model with emphasis. From the views of complexity, efficiency and forecast precision of models, the paper compared and analyzed the existing short-term traffic forecast methods, and determined to take the non-parameter regression theory as the foundation of model research. Based on the dynamic traffic data and the historical typical tendency database, the research calibrated the model parameters via the experimental study and Data tests, then, the multi-source historical tendency database based K-nearest neighbor non-parameter regression short-term traffic forecast model was proposed to predict average travel speeds up to 6 minutes into the future. Finally, the paper compiled the forecast procedure, several discretionary selected travel speed data series which collected from the FVDCS were used to validate the model. The results indicate that the model is capable of predicting average travel speeds with an accuracy of as high as above 90% by using the FCD, showing the feasibility and validity of the model. The model validation test also showed that the multi-source data will efficiently improve the forecasting precision and stability.However, because the coverage of real-time transportation data is limited, in order to forecast the traffic condition of entire road network, the paper also proposed the fuzzy forecasting technique of the road section which has no real-time traffic data provided. Through the characteristic analysis on the collected historical traffic data, the paper proposed one typical road network traffic pattern division standard for Beijing which takes the time period, weekday and the weather conditions into account. Then, the research validated the maximum likelihood traveling speed for various class roads under various traffic patterns, and constructed the fiducial forecasting fatabase.Finally, the traffic condition fuzzy forecast model was put forwarded, and it actualized the fuzzy forecasting of the road sections which have no real-time traffic data.Through researches in this paper, DVNS system achieved the processing and distribution of short-term traffic condition forecast information under the limited dynamic traffic data. By the unifying of the fuzzy forecast and the aggression forecast, the system provides the predicted information of the entire road network. This article will provides important reference for future researches based on the Floating vehicle data processing, the historical data processing, or short-term traffic forecast technique.